Tiny Mic: The #1 AI Mistake That's Slowing You Down (and How to Fix It)

AI is the smallest, biggest word that I hear used all the time. The one bit of advice that I have for our customers or for prospects is take a beat, take a minute to just understand the difference in architecture, technical architecture necessary for autonomous agents relative to perhaps where you've been experimenting with co pilots or chatbots. That in itself will help open your mind to why a platform approach may be the right way.

Hi, everyone, my name is Ivy Wright with Salesforce and I am here coming to you live from Fortune Brainstorm Tech in sunny Park City, Utah.

And I'm here with Shibani Ahuja, SVP Enterprise IT Strategy from Salesforce.

Thank you so much for being here with us today. We've sat in on a few panels, lots of great sessions so far. But I want to hear from you, from what you've sat in on so far, do you have any, like, big takeaways, Net new ideas? What are you hearing that you're like, I haven't thought about that.

I was on a panel yesterday and it was all about roi. And I loved the conversation about how different organizations are thinking about roi. Just, just the definition of it. It's so traditionally we think about, you know, cost cutting, capacity creation, but folks were looking at ROI in the context of higher quality data being the return on investment as you've got agents that can now audit their space or you've got growth as a, as an roi. So I just thought that was really great.

So I was in this session too, and I was of course taking notes. Same point. I love that there was someone that was talking about, we're able to pay our employees faster. And then another person had talked about, oh, it was you. You talked about auditing.

Yes.

You talked about this capability for AI agents to actually look in your background and audit. And that's how you found several mistakes even at Salesforce. Do you want to tell us about that?

So when we implemented agentforce on help.salesforce.com from what I had learned in the earlier implementation, we actually thought that agent was hallucinating, but the reality was it wasn't hallucinating. There was. The underlying information that we had available on our website was not duplicative, but it was conflicting. And so the agent actually helped us surface the fact that we've got knowledge articles out there that have different answers. So we turned that agent internally. That agent helped us to clean up our underlying data. And then when we pointed it back out, it was fabulous.

I mean, see right there, guys, Salesforce on Salesforce, this is what we're doing this is it.

We drink our own champagne.

But I did love, like, that whole conversation around different ways to find roi. And they were also saying on that panel that we almost need to, like, redefine how we're thinking about AI and even use new metrics. I thought that was such a good point.

We actually talked about a framework to say you can look at ROI in the context of efficiency, effectiveness, experience. And because I was an ex banker, Risk and control always got its own little section that's come up a lot, though.

Risk and control.

Risk and control has been topic of the day.

Yeah.

Security guardrails, auditability, traceability. All of it.

Yes. And I loved actually in our breakfast that we just had right now, someone had brought up the idea of a guardian agent.

Love that.

Okay, tell me more about what that is, how people can set this up, what its function looks like.

This idea of guardian agents was beautiful. It's this idea that it's not just the agents that are performing the actions, but you've got agents that are overseeing the agents that are performing the actions, and they're ensuring that there's integrity in how these agents are performing.

Okay, one more thing I loved about, like, thinking about agents in a different way. You said something that was like, such an aha moment for me. Okay, so you were talking about how think about agents as a hologram was the first point. But then you also talked about when we're thinking about agents more like humans, then there's kind of like this hierarchy of managers, and like they're on your team, like a person would be, and they report to this person. Okay, talk me through this, like, whole idea of how we should be thinking about agents.

So I'm going to tell you my origin story to Salesforce.

Yes, please.

How little I knew to how I sort of taught myself about what is this agentic architecture, Autonomous agents. So I joined Salesforce about 10 months ago, and I'd heard about Agentforce at that time because, you know, Salesforce is a marketing engine to be reckoned with. But in reality, 10 months ago, when I heard AgentForce, I thought, AgentForce, it's like a desktop for my contact center agents. Because at my previous organization, we called our agents the contact center agents. And I said, oh, yeah, and it's probably got a little thing to do with AI that's probably the Einstein embedded in. And it was an aha moment for me to say, wow, we need to slow it down. We need to actually describe what is an agent. So then I got jealous. I went, I Want my own billboard of an agent. And I started to think about this notion that there is a hologram of an agent that is sitting next to me. So we've got agent Astro instead of Piper. And I started to think about my onboarding to Salesforce and how I would relate it to that agent to really understand this notion. When I joined Salesforce, there was a job description that I had to apply for. An agent needs a job description. When I joined Salesforce, someone had to provision data, access to data. An agent needs the right provisioning of data, not access to all of the data.

Right.

When I joined, I was given metrics or KPIs or the actions that I need to perform in the role that I'm in. Similarly, an agent needs those parameters. When I joined, I needed a laptop or the channel in which I was going to engage. Maybe this, this is our channel where I'm engaging customers and colleagues. An agent needs a channel where it shows up and it can engage with customers, colleagues, like Slack. And then the last thing is, I usually have my access card and I go, when I joined Salesforce, someone said, you can open this door and you can't open this door. Agents need the parameters of which are the doors that you are allowed to open and not. And that's when it started to come to me to say, wow, this is like a hologram sitting next to me, a coworker. It's more than just, you know, question that I ask in a text box and get an answer and copy paste. Such a good analogy. It helped me understand it so well.

Yeah, no, such a good analogy. Shivani, you meet with a ton of CIOs, you're talking with them about their AI implementations and helping them with all of their roadmapping. What is like one or a few of the big misconceptions they have? What are they doing wrong? What is the thing that you're hearing over and over again?

The thing that they're doing wrong is building versus buying. They should be buying the platform.

There you go.

But biggest misconception is on Data Cloud, time and time again, every time Data Cloud comes up, I can practically read their mind. I volunteered. I'm like, look, Data Cloud is not another data lake or a data warehouse. I know that you are thinking that. I said, let me guess what's on your mind. Number one, you're thinking, I don't need another data lake, a data warehouse. I will never give you all of my data Salesforce. And three, you're probably thinking, I don't want to be locked in, because once you've got everything, then it's going to be really hard to exit. Yeah, that's what I thought when I first heard data cloud. And I didn't understand. And I'm like, I have Azure, I already have databricks, I have Snowflake. I don't need another one. Thank you. Until I started to talk to CIOs to say, how many people have a user ID and password to all of your data lakes and warehouses to extract that data near real time and activate it through your organization? Not one, because they aren't built for that operational performance. And so I said, data cloud is like a thin activation layer where you can harmonize the data. It still sits within your lakes and warehouses with zero copy. You're only accessing it as and when you need it, and you're able to activate it throughout your organization near real time.

Yeah.

And then that's when they go, huh, Interesting. I'm like, yes, this is what data cloud is.

Yeah. There was a gentleman from tiaa, he's their chief Information officer and he had brought up yesterday the fact that, you know, there are sexy use cases for AI, right? And then there are the big boring problems.

I love that. I love that he said that.

It literally woke me up at night. I wrote down on my phone, like in my notes, big boring problems.

Ivy, we need to talk about different things. If that's what's waking you up at night. This is. We needed those big boring problems.

They're keeping me up.

Boring problems.

Yeah. But he said this is where you really make the roi. It's really not the sexy use cases, it's solving those big boring problems.

Yes.

Yeah.

And I'll tell you, the big boring problems actually do two things. One is if you find a big boring problem that is like a big pain point for your colleagues and you're showing that through AI that you can address that and alleviate it, suddenly that becomes something desirable for colleagues. Going, did you just take away this administrative piece of thing that I had to do that, like, would keep me up at night? And you've taken that away, like, that becomes a win win.

Yeah, totally. In our breakfast that we just had, we were able to flash a little slide up on the screen, but you walked us through a four stage agentic maturity model. Tell me about this. Tell me about the different levels, how you move up. Tell me more.

This agentic maturity model is incredible. It's been created by a number of folks across the organization. It's like the culmination of the greatest minds coming together. This agentic maturity model, less about the levels it is about what it help our customers understand. Like one, it helps them get a common language for the biggest, smallest word, AI. People use AI so loosely, but AI can be everything from predictive to generative to autonomous or agentic. And The Maturity Model 1 helps people understand the plane of how you move from predictive to generative to autonomous. But it helps them understand where are they today and where do they aspire to be? Perhaps in the next year, in the next two years. It also helps them to cut through all the noise. Like so many customers say, I have 170 use cases. Like it's like overwhelm. I'm going take those use cases, break down the attributes of those use cases by the maturity model, tackle level one. That's it. Make only the tech investment in what you need to activate a singular use case. Harmonize only the data attributes that you need into data cloud to activate that singular use case. Validate that you're getting the returns and see that the value that you are getting from a consumption model is greater than what you would have might have been fearing before. And then move to level two where you're incrementally adding that architecture and then level three and level four.

Thank you for explaining that. Where do you think most CIOs are landing on that model today? And what's like maybe your one piece of advice for how they can move up to the next level.

So there is where they think they land on the maturity model people. I'm at level three, I'm at level four, I'm like, level four doesn't exist yet. We're almost there. A lot of CIOs would like to think that they're at level three. In reality, we're really looking at at best level one and level two use cases. In some cases they're talking about agentic architecture. What they're describing is a copilot.

Oh no.

And so it's very important that when you're in the room with CIOs that you really listen to the language that describes what their maturity is.

I know you had mentioned like a lot of times when you're in this meeting, you start by defining an AI agent and go from there.

Yeah, let's start with what's the implementation that you've had within your organization? Are you at predictive, Are you at generative or are you at autonomous? And that in itself strikes up the right conversation to say, well actually we're really looking at co pilots right now. And I said, great, so let's talk about agentic architecture. Because the architecture that you need for agentic agents to traverse the virtual halls of your organization is very different than what you're doing for a copilot.

Okay. So speaking of CIOs not knowing exactly where they're at and you know, some other failures, one thing that came up a lot was this idea of failure. And we've heard in the past, you know, fail fast, fail often. And it kind of felt like at Fortune, Brainstorm Tech that that is not the case. It feels like people are leaning in more to these kind of long failures where you learn more, you can adapt and actually these longer failures kind of turn into successes. So tell me a little bit about like how much time should we be allotting for failure?

You know, it's the, the notion fail fast, fail often. I think that it depends the time that you spend in the failure. I think that you stay in that failure as long as you can extract the learnings from it and from every angle, every point of view, that's how long you stay in that failure. Like we, we think about what we had talked about from our own implementation of an agent on help.salesforce.com, where it helped us identify anomalies in what we had in the underlying data that was always on our website.

Sure.

Immediately turn off that agent so that it's not continuing to give the wrong results. But then sit in that failure to say what happened? How can we extract value from this learning and you stay in that failure long enough to realize that maybe we need an agent that's going to look at our data and help us identify those anomalies.

I love that. Fail as long as it takes you to learn.

Exactly.

We can do that.

That's right.

Okay. Sam Altman recently called SaaS fast fashion and Satya Nadella said SaaS is dead. What is your gut feeling here? Where do you see SaaS industry heading from here? What's coming?

I love this. So SaaS is not dead, it's alive and well. I would say vertical SaaS, if you think about like single cloud solutions that may not be very sexy and very popular, but horizontal SaaS, where you've got platforms like Salesforce on a single code base that is now creating the infrastructure for people to agentic and autonomous architecture. I think about does a customer really want to build everything that is necessary to have digital agents or should they be looking at a partner that has a platform that enables it and maintains it and upkeeps it on their behalf. Like to me, horizontal SaaS is very well alive. Maybe vertical SaaS is taking a bit of a break.

Love that hot take. You were just on a panel this morning and I was listening in. They asked a question about, you know, where do you see technology going in the next five years? I want to ask you that same question because I absolutely loved your answer. So Shibani, what do you see happening in the next year in the AI world?

I was in Sydney not too long ago and I'm sitting there with the CIO and he asked me, shivani, like you're in this space, where do you think technology is going in the next five years? I like chuckled. I have no idea. Technology is moving so fast. And I said I would focus less on predicting where technology is going. I would focus more on not iq, not eq. Those are really well known on aq. Your adaptability quotient and that adaptability quotient comes from two places. One is your mindset of how are you adapting to the world as it's changing today and how technology is going to enable us. But the adaptability quotient of your tech stack, how interoperable, how open, how much can your stack play nicely with maybe the new emerging technologies that are coming? And I think of that in the context of Salesforce, where yes, we've got a very comprehensive platform, but that platform plays nicely with other technologies. So the adaptability quotient of your architecture is where I think people need to focus more.

Love that. I've got one more question for you and it is just around any advice you have for leaders that are starting out with AI, they're maybe mid journey. What's your best advice for them to move that up to the next level and succeed?

AI is the smallest, biggest word that I hear used all the time. The one bit of advice that I have for our customers or for prospects is take a beat, take a minute to just understand the difference in architecture, technical architecture necessary for a gentic architecture, for agents, for autonomous agents, relative to perhaps where you've been experimenting with copilots or chatbots. That in itself will help open your mind to why a platform approach may be the right way.